AlgoTrader
A production-grade automated US equity day-trading system. Screens the market pre-open, scores candidates with a LightGBM ML pipeline and FinBERT sentiment analysis, then executes 14 intraday algorithms through Alpaca's brokerage API, fully autonomous during market hours.
Challenge
Building a zero-intervention trading system that manages real capital across three float tiers, with robust risk controls, 4x PDT margin utilization, and reliable EOD liquidation, all on a cost-efficient serverless infrastructure.
Solution
Go trading engine on ECS Fargate with EventBridge scheduling (~81% cost reduction vs always-on), a 14-step risk approval pipeline, SQS FIFO for per-symbol ordering, weekly walk-forward ML retraining with SHAP pruning and KS drift detection, and a real-time Svelte dashboard via WebSocket push.
Technical Strategy
Go was chosen for the trading engine for its goroutine concurrency model and single-binary deployment, critical for processing 8 parallel symbol streams with sub-millisecond latency. Python handles the ML layer (LightGBM, FinBERT) where ecosystem maturity matters more than speed. ECS Fargate with EventBridge scheduling eliminates idle compute costs, while SQS FIFO guarantees per-symbol message ordering without a self-managed broker. Terraform codifies the entire stack for reproducible deploys.